metadata
base_model:
- mistralai/Ministral-8B-Instruct-2410
This is the 8-bit quantized model of the ministral-8B by Mistral-AI.Please follow the following instruction to run the model on your device:
There are multiple ways to infer the model. Firstly, let's install llama.cpp and use it for the inference:
- Install
git clone https://github.com/ggerganov/llama.cpp
!mkdir llama.cpp/build && cd llama.cpp/build && cmake .. && cmake --build . --config Release
- Inference
./llama.cpp/build/bin/llama-cli -m ./ministral-8b_Q8_0.gguf -cnv -p "You are a helpful assistant"
Here, you can interact with model from your terminal.
Alternatively, we can use python binding of the llama.cpp to run the model on both CPU and GPU.
- Install
pip install --no-cache-dir llama-cpp-python==0.2.85 --extra-index-url https://abetlen.github.io/llama-cpp-python/whl/cu122
- Inference on CPU
from llama_cpp import Llama
model_path = "./ministral-8b_Q8_0.gguf"
llm = Llama(model_path=model_path, n_threads=8, verbose=False)
prompt = "What should I do when my eyes are dry?"
output = llm(
prompt=f"<|user|>\n{prompt}<|end|>\n<|assistant|>",
max_tokens=4096,
stop=["<|end|>"],
echo=False, # Whether to echo the prompt
)
print(output)
- Inference on GPU
from llama_cpp import Llama
model_path = "./ministral-8b_Q8_0.gguf"
llm = Llama(model_path=model_path, n_threads=8, n_gpu_layers=-1, verbose=False)
prompt = "What should I do when my eyes are dry?"
output = llm(
prompt=f"<|user|>\n{prompt}<|end|>\n<|assistant|>",
max_tokens=4096,
stop=["<|end|>"],
echo=False, # Whether to echo the prompt
)
print(output)